機械学習によるマルチスケール物理シミュレーションの高度化

2022 Fiscal Year Annual Research Report

• Project/Area Number: 22F22708
• Allocation Type: Single-year Grants
• Research Institution: Kyoto University
• Principal Investigator: 鹿島 久嗣 京都大学, 情報学研究科, 教授 (80545583)
• Co-Investigator(Kenkyū-buntansha): BARBOT ARMAND 京都大学, 情報学研究科, 外国人特別研究員
• Project Period (FY): 2022-09-28 – 2025-03-31
• Keywords: マテリアルズインフォマティクス / 機械学習 / 人工知能

Outline of Annual Research Achievements

This research project aims at implementing nucleation of dislocation in mesoscopic scale simulation by training a machine learning model with data obtained from simulations at the atomistic scale. Since the beginning of the project, seven months ago, the fellow was able to train a model able to predict with a very high degree of precision the nucleation of dislocation in atomistic simulations by training a machine learning model with : (a) the shape of the system, (b) the strain, (c) the global potential energy, (d) at which strain interval took place the last plastic event, and (e) the size of the system. This result shows that it is indeed possible to use machine learning to predict nucleation of dislocations and allows to start the implementation of the model at the mesoscopic scale.

Current Status of Research Progress

2: Research has progressed on the whole more than it was originally planned.

Reason

プロジェクト立ち上げのための調査や協力機関との議論を通じて、順調なスタートを切ったといえる。

Strategy for Future Research Activity

As the trained machine learning model is able to predict nucleation of dislocation with a high precision, the fellow will start the implementation of this model at the mesoscopic scale by working collaborating lab in France which is specialized in mesoscopic simulations. This task will occupy the following months and will be published in a peer-reviewed journal. In parallel, the fellow is also considering another approach using Physically Informed Neural Networks (PINN) to obtain a more cost-efficient model. Finally, if the implementation of the nucleation of dislocation is successful, the fellow will work on developing new machine learning based models to implement other plastic phenomena at the mesoscopic scale from atomistic simulation, such as the cross-slips.